AI is spitting out more potential drugs than ever. This startup wants to figure out which ones matter. | TechCrunch

Open original

10x Science built a platform that uses chemistry-based algorithms and AI agents to interpret complex mass spectrometry data, speeding up protein characterization for drug development. Backed by a $4.8 million seed round, it helps biotechs and pharma quickly validate AI-generated treatment candidates. The startup plans to refine its models and expand offerings by integrating broader cellular data.

  • 10x Science raised a $4.8M seed round led by Initialized Capital to build AI that interprets mass spectrometry data for drug candidate validation, addressing the bottleneck created by AI protein-folding tools flooding pipelines with candidates.
  • Its platform is fully traceable rather than a black box, which matters for regulatory compliance—unlike prior tools that over-promised or failed on complex molecules.
  • Early users like Rilas Technologies report it shaves weeks off workflows, with the AI accurately identifying proteins by name and pulling sequences from public databases.
  • Long-term plan is to layer cellular data on top of protein structure to create "molecular intelligence," positioning it as a potential go-to analytics engine for AI-driven drug development.

Google DeepMind’s protein-folding breakthroughs have flooded biopharma with new drug candidates, but testing each one remains slow and costly. That’s where 10x Science comes in. Launched in December 2025 by chemical biologist David Roberts, biologist Andrew Reiter and AI specialist Vishnu Tejus, the startup just closed a $4.8 million seed round led by Initialized Capital (with YC, Civilization Ventures and Founder Factor joining). Their platform merges chemistry-driven algorithms and AI “agents” trained on mass-spectrometry data to turn raw spectra into clear molecular structures. Unlike black-box models, 10x Science can trace every step of its analysis—a must for regulatory compliance. Scientists at Rilas Technologies have run dozens of samples through the platform and report it shaves weeks off their workflows. Matthew Crawford, one of their users, says the AI not only interprets spectra accurately but even infers which protein he’s testing by name and then pulls the sequence from public databases. Previous tools often over-promised or tripped over complex molecules, but 10x Science nails reasonable assumptions, thanks to the founders’ hands-on lab experience in Carolyn Bertozzi’s Stanford group. With seed proceeds in hand, 10x Science plans to hire more engineers, refine its models and onboard pharma clients and academic labs. In the short term, it’s a subscription-based SaaS that pharmaceutical companies must pay to process candidates through the “funnel.” Longer term, Roberts aims to layer cell-level data on top of protein structure to build what he calls “molecular intelligence.” Investors like Initialized’s Zoe Perret see a defensible niche: few teams understand both deep biochemistry and the messy data mass spectrometry produces. If the startup scales, it could become a go-to analytics engine for anyone converting AI-generated candidates into real-world drugs.

Ask about this article